移动群智感知多任务参与者优选方法研究

Multitask-Oriented Participant Selection in Mobile Crowd Sensing

该文主要研究面向移动群智感知的任务分发方法,提出了一种新的参与者选择:面向多任务并发的参与者优选.不同于其他参与者选择,文中选择出的参与者不再局限于只能完成一个任务,每个参与者可以在规定时间内尽可能的完成多个任务,由此降低群智平台的成本.并提出MultiTasker方法,其目标是选择出最佳的参与者集合,使参与者完成任务所移动的总距离最短以降低成本,并且完成任务的参与者人数最少以优化用户资源.为了实现这个目标,文中设计了3种算法:T-Random、T-Most和PT-Most.T-Random和T-Most是以任务为中心进行参与者选择,而PT-Most是以用户为中心进行参与者选择.通过一个大规模的真实数据集对设计的3种算法进行实验评估,同时研究了参与者选择情况与各种因素之间的关系,如任务分布和任务执行时间等.

Participant selection or task allocation is a key issue in Mobile Crowd Sensing（MCS）.While previous participant selection approaches mainly focus on selecting aproper subset of users for a single MCS task,multitask-oriented participant selection is essential and useful for the efficiency of large-scale MCS platforms.This paper studies the task allocation issue in MCS,and we propose a new multitask-oriented participant selection problem.The difference between our work and other studies in participant selection is that each participant can complete as many tasks as possible within the given time period,rather than merely one task.This participant selection problem has some advantages：First,each participant can complete multiple tasks,and this will bring benefits to the running of the MCS platform.It is particularly useful in cases that user resources are not sufficient regarding to the published tasks.Second,there exist geographical proximity among the published tasks,and we can optimize the MCS platform performance by assigning the tasks as a whole.Third,each participant can perform as many tasks as possible,and this may improve personal income and participation enthusiasm.To address this participant selection issue,we propose MultiTasker,the objective of which is to minimize the total distance the selected participants move while minimizing the number of participants,subjecting to that all tasks are allocated with the needed number of participants and the tasks are completed by each selected participant within the given time period.In order to achieve the objective,we propose three algorithms：T-Random,T-Most and PT-Most.T-Random and T-Most select participantsin a task-centric way,while PT-Most selects participants in a people-centric manner.We evaluated the three algorithms using a large-scale real-world dataset,and studied the relationship between participants and other factors,such as the distribution of tasks and task completion time.